Research Of Iterative Dynamic Programming For Industrial Process

Steady-state during the complex industrial process is often temporary and relative, because of the various fluctuations, disturbances and condition changes in practice. While the traditional methods in chemistry industrial process are difficult to deal with those problems caused by dynamic changes, dynamic optimization has become very urgent in academic and industry research due to its excellent performance in dealing with dynamic changes.In this paper, some explorations on the basis of the current research status are developed on three aspects:the improvements for solving differential equations during iterative dynamic programming process, state constraint-first strategy for boundary fixed problems based on iterative dynamic programming and an improved slack variable approach effectively for dynamic processes with inequality state constraints. After the verification on a series of classic benchmarks of industrial process dynamic optimization, the above work play a visible role in the solution accuracy, convergence speed or stability effect. To be specific, the main work and contributions are as follows:(1) Since the solving process of iterative dynamic programming involves a variety of differential equations, the speed of solving equation would largely affect the overall operation efficiency. In this paper, several improvements are proposed to attempt to raise the rate of equation running, including improved fourth-order Runge-Kutta algorithm, Euler method and multi-step Adams algorithm. Conclusions show that the fourth-order Runge-Kutta can achieve the balanced effect between the accuracy and running speed, while, the improved Euler algorithm can reach the better results but with longer time, and Adams algorithm would in more degree get the excellent optimization accuracy and efficiency.(2) Aiming at the dynamic optimization problem with final value constraints, a state constraint-first strategy based on iterative dynamic programming is proposed, where, the performance goal and constraints goal are divided into separate two individual optimization problems. After the constraints goal is achieved, the feasible solutions are substituted into the performance function to search the best control variable. Then the optimal control trajectories and optimal states trajectories which satisfy the both two goals could be found. By the classic benchmarks of industrial process dynamic optimization, the research results are proved to be efficient and have obvious advantages.(3) As to the dynamic optimization problem with inequality state constraints, an improved slack variable approach is proposed. By introducing a specific slack factor, the original problem is turned into the equivalent unconstrained equations, then the optimal control and states trajectories are searched by iterative dynamic programming. Several classic benchmarks of industrial process dynamic optimization are used to test and verify the good control and optimization effects of the proposed approach and the research results reveal its effectiveness.